Brad DeLong -In Sunday's Gallup Sample It Was Obama (D) 57% Romney ($) 38%

[DemocratSinceBirth]

IT LOOKS LIKE GALLUP'S 10/7 POLLING WAS 57 OBAMA, 38 ROMNEY...

The most recent Gallup Presidential Election Trial Heat Results: Barack Obama vs. Mitt Romney was a very, very strong day for Obama:

Gallup reports:

Gallup October 1-7: Obama 50%, Romney 45%
Gallup October 4-6: Obama 47%, Romney 47%
Gallup September 30-October 2: Obama 50%, Romney 45%

Since Gallup on October 6--the span that includes 9/30-10/2 and 10/4-10/6--was 49-46, that tells us that the October 3 results were roughly 52-46 Obama-Romney…

Since Obama gained from October 6 to October 7, when 9/30 dropped out and 10/7 was added, that tells us that 10/7 was stronger for Obama and weaker for Romney than 9/30. If 9/30 was the average for 9/30-10/2--50-45--that tells us that…

October 7 was roughly 57-38 Obama-Romney!

(If 9/30 was weaker than 50-45 for Obama and 10/1 and 10/2 were stronger, 10/7 was less of a pro-Obama signal. If 9/30 was stronger and 10/1 and 10/2 weaker than 50-45, then 10/7 was an even stronger pro-Obama signal.)

This is why Gallup does not release its daily tracking poll results--the sample sizes are so small that there is so much statistical noise in them that they are of little value...


http://delong.typepad.com/sdj/2012/10/it-looks-like-gallups-107-polling-was-57-obama-38-romney.html

[hrmjustin]

1. I wish that they would publish the day to day numbers.

[cthulu2016]

2. We cannot know that

Last edited Mon Oct 8, 2012, 04:23 PM - Edit history (3)

we know that would be the case if 9/30 was 50-45, but we don't know what 9/30 was.

So 10/7 cannot be calculated.

I was probably a very good day for Obama.

[regnaD kciN]

3. Exactly...

I've been doing analysis of the tracking poll movements post-debate. In such analysis, there are some things you can calculate, and some you can't. The best you can do is figure out roughly how much of a shift took place between the latest sample and the one just discarded. But you cannot assign absolute numbers to each day, without knowing the results from each of the initial seven days. Even then, rounding errors dictate that each day thereafter is an approximation -- and, the further forward you go, the rougher that approximation goes.

[DemocratSinceBirth]

4. Gallup Was Too Cute By Half By Publicizing A Three Day Sample

They encouraged this kind of manipulation of the data.

[woolldog]

7. I don't get why they would do such a thing.

Clearly sunday was very good for Obama. But they don't mention that. I'm going to have to start entertaining the idea that Gallup is agenda driven. There's no other reason to manipulate the data like this.

[gcomeau]

5. No It Was Not

I'd love that to be the case as much as anyone but I can't turn a blind eye to crappy math.

Without knowing what the single day results from 9/30 were... which we *do not know*, we cannot know how good Sunday's were, since the movement in the 7 day average is the combination of dropping 9/30 and adding 10/7.

All we know is 10/7 was better than 9/30. 9/30 could have been pretty crappy (by pre-debate standards) and Sunday was ok. Or 9/30 it could have been pretty good and Sunday was fantastic. But there is zero way to tell that from the available data. At best you can make vague guesses based on how the 7 day average was moving over the preceding time.

Can we please stop getting threads saying Obama had some ridiculous +12 or +14 or +19 day on Sunday? The odds against that actually happening are astronomical. He likely had a *good* day on Sunday. Take it and be happy with the reality.

[DemocratSinceBirth]

6. By Gallup Releasing Two Tracking Polls On The Same Day They Encouraged These Machinations

The onus is on them and not some random bloggers.

[gcomeau]

8. They did not release two tracking polls.

They released the exact same 7 day poll they always release.

They then also shared details of what some of the isolated 3 day sections *inside* that 7 day poll looked like, which they normally don't tell people. And everyone went crazy thinking that was some separate poll.